Meta's newest AI discovers stronger and greener concrete formulas
The results were stronger than the index while using 40 percent less carbon.
They may not be able to shout "Eureka!" like their human colleagues but AI/ML system have shown immense potential in the field of compound discovery — whether that's sifting through reams of data to find new therapeutic compounds or imagining new recipes using the ingredients' flavor profiles. Now a team from Meta AI, working with researchers at the University of Illinois, Urbana-Champaign, have created an AI that can devise and refine formulas for increasingly high-strength, low-carbon concrete.
Traditional methods for creating concrete, of which we produce billions of tons every year, are far from ecologically friendly. In fact, they generate an estimated 8 percent of the annual global carbon dioxide emission total. Advances have been made in recent years to reduce the concrete industry's carbon footprint (as well as in make the material more rugged, more resilient and even capable of charging EVs) but overall its production remains among the most carbon intensive in modern construction.
Reducing the amount of carbon that goes into concrete could be as simple as changing the ingredients that go into concrete. The material is made from four basic components: cement, aggregate, water and admixture (which act as doping agents). Cement is far and away the most carbon-intensive ingredient of the four so research has been made into reducing the amount of cement needed by supplementing it with lower-carbon materials like fly ash, slag, or ground glass.
Similarly, aggregate materials like gravel, crushed stone, sand might be replaced with recycled concrete. The problem is that there are dozens of potential ingredient materials that could be used and the ratio of their amounts all interact to influence the structural profile of the resulting concrete. In short, there are a whole slew of possible combinations for researchers to test, select, and refine; and working through those myriad options sequentially, at human speed, is going to take forever. So the Meta folks trained an AI to do it, much faster.
Working with Prof. Lav Varshney, electrical and computer engineering department, and Prof. Nishant Garg, civil engineering department, both of the University of Illinois at Urbana-Champaign, the team first trained the model using the Concrete Compressive Strength data set. This set includes more than 1,000 concrete formulas as well as their structural attributes, including seven-day and 28-day compressive strength data. The team determined the resulting concrete mixture's carbon footprint using the Cement Sustainability Initiative's Environmental Product Declaration (EPD) tool.
Of the generated list of potential formulas, the research team then selected the five most promising options and iteratively refined them until they met or exceeded the 7- and 28-day strength metrics while dropping carbon requirements by at least 40 percent. The refinement process took mere weeks and ended up generating a concrete formula that exceeded all of those requirements while replacing as much as 50 percent of the required cement with fly ash and slag. Meta then teamed with concrete company Ozinga, the folks who recently built Meta's newest datacenter in Illinois, to further refine the formula and conduct real world testing.
Looking ahead, the Meta team hopes to further improve the formula's 3- and 5-day strength profiles (basically ensuring it dries faster so the rest of the construction can move ahead sooner) and get a better understanding of how it cures under varying weather conditions like wind or high humidity.